Multivariate small area estimation of undernutrition for children under five using official statistics

Surveys are mainly used to obtain reliable estimates for planned domains at national and regional levels. However, the unplanned domains (lower administrative layers) with small sample sizes must be estimated. The direct survey estimates of the non-planned domains with small sample sizes lead to large sampling variability. Thus, small area estimations dealt with managing this variability by borrowing the strength of neighboring areas. The target variables of the study were obtained from the 2016 Ethiopian demographic and health survey (EDHS) and the auxiliary variables taken from the 2007 population and housing census data. Multivariate Fay Herriot (MFH) model was used by incorporating the correlations among the target variables. The model diagnostic measures assured the normality assumption, and the consistency of multivariate small area estimates are valid. Multivariate EBLUPs of the target variables produced the lowest percent coefficient of variation (CV) and root mean square error (MSE). Therefore, multivariate EBLUP has improved the direct survey estimates of undernutrition (stunting, wasting, and underweight) for small sample sizes (even zero sample sizes). It also provided better estimates compared to the univariate EBLUPs. Generally, multivariate EBLUPs of undernutrition produced the best reliable, efficient, and precise estimates for small sample sizes in all zones. Zones are essential domains for planning and monitoring purposes in the country, and therefore these results provide valuable estimates for policymakers, planners, and legislative organs of the government. One of the novelties of this paper is estimating the non-sampled zones, and therefore the policymakers will give equal attention similar to the sampled zones.